Quantitative MRI in refractory temporal lobe epilepsy: relationship with surgical outcomes

Abstract

Medically intractable temporal lobe epilepsy (TLE) remains a serious health problem. Across treatment centers, up to 40% of patients with TLE will continue to experience persistent postoperative seizures at 2-year follow-up. It is unknown why such a large number of patients continue to experience seizures despite being suitable candidates for resective surgery. Preoperative quantitative MRI techniques may provide useful information on why some patients continue to experience disabling seizures, and may have the potential to develop prognostic markers of surgical outcome. In this article, we provide an overview of how quantitative MRI morphometric and diffusion tensor imaging (DTI) data have improved the understanding of brain structural alterations in patients with refractory TLE. We subsequently review the studies that have applied quantitative structural imaging techniques to identify the neuroanatomical factors that are most strongly related to a poor postoperative prognosis. In summary, quantitative imaging studies strongly suggest that TLE is a disorder affecting a network of neurobiological systems, characterized by multiple and inter-related limbic and extra-limbic network abnormalities. The relationship between brain alterations and postoperative outcome are less consistent, but there is emerging evidence suggesting that seizures are less likely to remit with surgery when presurgical abnormalities are observed in the connectivity supporting brain regions serving as network nodes located outside the resected temporal lobe. Future work, possibly harnessing the potential from multimodal imaging approaches, may further elucidate the etiology of persistent postoperative seizures in patients with refractory TLE. Furthermore, quantitative imaging techniques may be explored to provide individualized measures of postoperative seizure freedom outcome.

Introduction

Medically refractory temporal lobe epilepsy (TLE) is the most common and most frequently operated intractable form of epilepsy (1,2). Resective temporal lobe surgery is effective for many patients with refractory TLE, providing a greater likelihood of seizure freedom and improved quality of life compared to anti-epileptic drug (AED) treatment for suitable patients (3). However, up to 40% of patients with refractory TLE will continue to experience disabling postoperative seizures 2 years after surgery (4-7), and the proportion continuing to have seizures increases with longer postoperative follow-up (8,9). It is currently unknown why a large subgroup of patients continues to experience postoperative seizures despite surgical intervention. Prior to surgery, the patients who achieve complete postoperative seizure control are typically clinically indistinguishable from patients who continue to experience seizures and there is considerable controversy in the current literature regarding presurgical clinical factors that may help predict outcome. Prognostic stratification of patients according to likely outcome is therefore very difficult based on the manifestation of seizures and natural history of the disorder.

Outcome predictors of TLE surgery have been extensively debated in the literature over the past decades. Frequently reported indicators of optimal surgical outcome are unilateral signs of HS on diagnostic MRI (10-13) and unilateral interictal epileptiform discharges (13-15). Other reported predictors of good outcome are younger age at surgery (16), shorter epilepsy duration (16), history febrile seizures (17,18), epileptiform discharge frequency (19), and absence of generalized seizures (20,21).

It is important to note that these indicators are not consistently observed across studies, even among large well-powered analyses. Furthermore, despite that presentation of HS on MRI is most strongly related to an improved outcome, it does not guarantee an optimal one; at 60 months after surgery, only half of all patients with TLE and HS will experience no postoperative seizures (11). Table 1 provides a summary of findings from a sample of moderate to large studies addressing factors predicting seizure outcome after TLE surgery. It is noteworthy that the results from various studies are partially discordant; some features are considered predictive of seizure freedom after surgery in some studies but not in others. While it is possible that specific features may have not been identified as predictors in some studies due to lack of statistical power, it is also equally possible that some features are not predictors for all patients with TLE; in other words, they may not be generalizable to samples of patients outside the specific studies.

Table 1. Results from selected moderate and large observational studies addressing factors influencing seizure outcome after temporal lobe epilepsy surgery.

Study	No. of TLE patients studied	Predictors of surgical seizure outcome
Berkovic et al. (11)	135	Absence of unilateral HS or other lesions on preoperative MRI was associated with suboptimal post-operative seizure control
Janszky et al. (18)	133	History of complex febrile convulsions was associated with better surgical prognosis when compared with no history of febrile convulsions
Jeong et al. (20)	277	Younger age at surgery, absence of history of generalized tonic-clonic seizures and HS on MRI were associated with good surgical outcome
Kilpatrick et al. (22)	56	Clinical factors such as seizure frequency and duration of epilepsy were not associated with surgical outcome
Krendl et al. (19)	55	A higher frequency of interictal epileptiform discharges was associated with worse postoperative seizure control
Radhakrishnan et al. (13)	175	Preoperative evidence of unilateral HS, or of interictal epileptiform discharges (concordant with the location of ictal onset) were associated with excellent outcome
Schulz et al. (15)	58	Contralateral ictal propagation or interictal epileptiform discharges were associated with worse outcomes
Spencer et al. (21)	297	Hippocampal atrophy and a history of absence of generalized tonic-clonic seizures were the sole predictors of 2-year remission

Overall, the data above demonstrates two important points. First, although unilateral HS appears to be the most reproducible factor leading to optimal outcome, many patients with unilateral HS do not achieve seizure freedom. Second, given the numerous high-quality (single and multi-center) outcome studies performed to date, it is highly unlikely that new discoveries about prognosis can be gained by further studies assessing typical preoperative clinical factors. These points underscore the crucial need for a novel biomarker to predict outcome.

The goal of this paper is to provide a review of the studies that have attempted to determine the preoperative quantitative MRI correlates of postoperative seizure outcome in patients with refractory TLE. The identification of preoperative imaging correlates of persistent postoperative seizures may lead to the development of novel biomarkers of treatment outcome for patients with TLE. We begin by providing an overview of brain alterations in TLE before focusing on studies that have directly examined postoperative outcome, and attempt to resolve whether preoperative quantitative structural alterations have any significance for the prediction of persistent postoperative seizures.

Quantitative MRI in TLE: overview of structural alterations

Volumetric MRI

Quantitative MRI techniques (e.g., hippocampal volumetry) are occasionally used to provide supplementary diagnostic information in context of preoperative evaluation (23,24), and have been widely used to characterize alterations in brain structure in patients with TLE. There are several articles that provided early reports of the reliable detection of hippocampal atrophy (HA) in patients with TLE based on conventional region-of-interest volumetry applied to (typically T1-weighted) MR images as a surrogate marker of HS (25-30). The application of similar MRI methods to other brain structures later revealed extrahippocampal atrophy, particularly of limbic and paralimbic regions (31,32). Atrophy was reported of regions adjacent to and closely connected with the hippocampus, including the amygdala, and entorhinal and perirhinal cortices (33-36), Zand of lateral neocortical temporal lobe regions (37) preferentially, but not exclusively, to the side of seizure onset. Structural alterations were also reported of deep grey matter nuclei known to be important for the modulation, propagation and expression of focal seizures, including the thalamus and striatum (38-40). Whether these changes are the result of recurrent uncontrolled seizures or are pre-existing is of considerable debate (41). Some studies have revealed a significant relationship between the duration of TLE and hippocampal (42-45) and extrahippocampal (39,46) volume, and between estimated number of seizures and volume (47,48), suggesting that the chronicity of the disorder, potentially including the excitotoxic effects of recurrent seizures, has a pathologically degenerating effect on the brain. Other cross-sectional studies reported no relationships between volume and duration of TLE (49,50). Some longitudinal studies have reported subtle progressive volume loss of the presumed epileptogenic hippocampus (51-53) and extrahippocampal cortex (54) across cohorts of patients with TLE, whilst others have not (55). The observed inconsistency in volumetric MRI studies on the progression of brain atrophy and the etiology of HS in refractory TLE mirrors that seen in experimental animal studies of induced epilepsy (56).

One important limitation associated with manual morphometry techniques is with respect to their limited applicability in clinical practice. They require a time-consuming process by a skilled rater that is usually impractical in busy clinical centers. Furthermore, the reliability and reproducibility of the method should be periodically checked since the results can be dependent on the rater, posing significant problems if there is a change in staff (e.g., if the rater who assessed the normative sample of controls is not the same as the rater who tests patients). For these reasons, a clinically useful biomarker should rely on automated, rater independent measures that are biologically substantiated.

Developments in MRI analysis techniques throughout the late 1990s and 2000s permitted the automated quantitative analysis of brain structure not restricted to one region-of-interest, overcoming the time and resource consuming nature of human volumetric measurements. The most widely applied automated MRI analysis technique applied to refractory TLE is voxel-based morphometry (VBM) (57). To our knowledge, there are five review articles currently published on the application of VBM studies to study brain alterations in refractory TLE (58-62), and the reader is referred to these papers for a detailed review of studies. Generally, VBM studies have revealed a topological pattern of brain atrophy relatively consistent with earlier region-of-interest volumetric studies, indicating a primary limbic and paralimbic distribution of atrophy. The most common structural alteration revealed by VBM is ipsilateral hippocampal and thalamic atrophy (32,58,60-62). The primary area of the thalamus affected appears to be the medial dorsal area (62) (Figure 1), which is a region of the thalamus known to be reciprocally connected with the temporal lobe (63) and anatomically and physiologically altered in animal models of limbic epilepsy (64,65). There is inconsistent evidence on the effect of epilepsy duration on hippocampal grey matter volume in VBM studies (66-69), but more consistent evidence indicating a relationship between increasing thalamic atrophy and epilepsy duration (32,66,67,69). Automated quantitative techniques that permit the analysis of cortical thickness over the whole brain (Figure 2) have revealed bilateral medial and lateral multi-lobar cortical atrophy in patients with unilateral TLE, with some variation in the extent and topological distribution of atrophy (71-79).

Figure 1.

Hippocampal and medial dorsal thalamic atrophy based on a meta-analysis of VBM studies of TLE (62). Significant areas of convergence in a VBM-anatomic likelihood estimation analysis in rectified (A), right TLE (B), and left TLE (C) groups. Meta-analytic connectivity modeling analyses (D) of left (green) and right (red) hippocampal tissue labels indicating co-activation in the medial dorsal thalamic region. Figure reproduced with author’s permission. VBM, voxel-based morphometry; TLE, temporal lobe epilepsy.

Figure 2.

Significant differences in regional cortical thickness in patients with left and right mTLE relative to healthy controls as reported in one study (70). Similar to other studies, cortical thinning—a proxy for atrophy—is observed in multi-lobar cerebral regions distributed bilaterally across the brain. mTLE, mesial temporal lobe epilepsy.

One significant issue with respect to the interpretation of whole-brain voxel based morphometric MRI changes is that there is still insufficient information on the histopathological basis of extrahippocampal, particularly extratemporal lobe structural alterations in TLE. As discussed later, it is important to understand the underlying pathology of MRI changes in TLE so that the biological etiology of postoperative seizures can be resolved. It is known that hippocampal alterations on MRI correlate with histopathologically quantified neuronal loss given that resected hippocampal material can be readily analysed and quantitatively related to preoperative hippocampal volume (27,30,80-83). Similarly, temporal lobe neocortex can be histopathologically examined in patients who undergo anterior temporal lobectomy (ATL). Results have indicated a wider range of lateral temporal lobe neocortical pathology, including gliosis and the presence of heterotopic white matter neurons in some patients with presumed mesial TLE (84). A recent post-mortem study reported significantly reduced neuronal density in the mediodorsal thalamic nucleus in patients with TLE ipsilateral to the side of HS (85), which is consistent with the VBM meta-analysis discussed above (62). However, the inference that MRI-determined grey matter volume loss is a proxy for pathological neuronal loss within cortical regions outside the temporal lobe is currently unsubstantiated.

Given the recent modifications in the classification of epilepsy disorders to consider the importance of brain networks involved in seizure onset, including focal epilepsies (86,87), there has been a new direction of research in TLE to model neuroimaging data in terms of connectivity networks (88). Although this is most eloquently done through analysis of diffusion tensor imaging (DTI) data (see below), inferential analysis of brain structural networks in TLE has been performed using structural MRI covariance, particularly by applying graph theory to segmented and parcellated cortical and subcortical regions-of-interest or correlations in the size, volume or thickness between a single seed region and multiple target regions (79,89-91). For example, Bernhardt et al. evaluated the topographical patterns of cortical atrophy in patients with TLE and observed that, in comparison with controls, patients demonstrated various abnormalities in the network formed by the covariance of regional brain volumes (79). Specifically, patients with epilepsy demonstrated a reorganization of network hubs, as well as changes in global network configuration. Bonilha et al. (92) adopted a similar methodological approach in a recent study examining the organization of cortical covariance networks in children with recently diagnosed epilepsy. This study revealed a similar pattern of topological changes in networks in children with epilepsy, notably due to a significant rearrangement of the regional distribution of network hubs.

One important limitation associated with cortical covariance networks is the fact that only a single network can be determined across a group of subjects. Specifically, one has to assess the covariance across pairs of regions as determined based on the distribution of volumes where each subject is a single data point. Subsequent statistical analyses are then performed based on the putative distribution of data from resampled strategies. Therefore, this method does not permit the assessment of neural architecture at single individual level. Instead, it provides an overview of network abnormalities for TLE as a group.

Diffusion tensor imaging (DTI)

Even though grey matter neurons are presumed to be the generator of seizure activity, white matter is an integral part of the epileptogenic network since axons are the transmission pathways of the brain (93), thus providing the framework for seizure onset and propagation. DTI can provide inferential analysis of neuronal connectivity and structural networks through measurement of water diffusion in the brain, permitting a window into microscopic alterations in patients with epilepsy. Similarly to studies using volumetric MRI, there are many applications of DTI techniques to study brain alterations in refractory TLE. Analysis techniques can be largely categorised into four types: (I) quantification of voxel-based DTI scalar values [e.g., fractional anisotropy (FA) and mean diffusivity (MD)] across the whole brain or within predefined regions-of-interest (e.g., within hippocampus, temporal lobe white matter); (II) probabilistic tractography that generates probabilistic white matter paths between a seed and target region-of-interest; (III) deterministic tractography that permits manual reconstruction of known white matter tracts; and (IV) connectome approaches that typically use tractography analyses to build models of whole brain structural networks and connectivity. A detailed review of DTI studies in refractory TLE is beyond the scope of this article, and the reader is referred to other sources (88,89,93,94). As reviewed by Bernhardt et al. (89), DTI studies consistently show decreased FA in temporo-limbic white matter tracts in groups of patients with TLE, including the fornix, parahippocampal fibers, uncinate fasciculus and cingulum bundles, and in more widespread regions, including the inferior and superior longitudinal fascicles, the internal and external capsules and the corpus callosum. Alterations in tract MD alterations appear more restricted and decrease as a function of anatomical distance to the temporal lobe (89,95). Deep grey matter nuclei also show evidence of diffusion abnormalities, primarily manifested as reduced FA and increased MD of the thalamus (96-99), and an increase in FA of the putamen (96,97).

The opportunity afforded by quantitative DTI to analyse properties of large-scale brain connectivity networks has led to an emerging field in neuroimaging connectomics, which is increasingly being applied to understand the anatomical and physiological basis of TLE (94). Graph theory is applied to quantify and compare properties of networks constructed from nodes (typically multiple cortical and subcortical regions-of-interest obtained from parcellated T1-weighted MRIs) and edges (in the case of DTI for structural connectivity, connecting tracts) distributed across the brain. This is an emergent field within neuroscience and it is particularly promising for epilepsy research, as epilepsy is traditionally considered to be a disease related to abnormal brain networks. Indeed, the neurobiological mechanisms associated with epileptogenesis are tightly linked with aberrant neuronal connectivity.

The concept of epilepsy as a network disease has gained popularity over the past few decades (88,100). In fact, as mentioned above, the notion of epilepsy as a network process has guided the revision of the Classification of Seizures and Epileptic Syndromes (86). This is particularly relevant in the context of TLE since accumulating evidence suggests that extrahippocampal pathology is present in TLE and may configure a network of abnormal structures involved in the generation of seizures (101). Studies employing quantitative imaging methods have consistently demonstrated that patients with TLE exhibit a pattern of structural abnormalities that, albeit invisible on visual inspection of MRI, involve brain structures beyond the hippocampus and the temporal lobe (31-33,58,68,73,102,103). Hence, abnormal extrahippocampal regions can constitute an abnormal network, which may originate and maintain seizures after the removal of the medial temporal lobe and lead to postoperative seizure recurrence (100). Until now, limitations in brain mapping technology have prevented the accurate assessment of individualized patterns of abnormal networks to test the hypothesis suggesting that neural network architecture is associated with surgical responsiveness.

The number of articles demonstrating abnormal connectomes in patients with TLE is still limited given the recent development of the technology. Nonetheless, there is cumulating evidence suggesting that epilepsy, and in particular TLE, is associated with vast connectome abnormalities. For example, Bonilha et al. who pioneered this field, demonstrated that TLE is associated with such abnormalities, notably increased limbic network clustering and efficiency in spite of regional fiber loss (104). DeSalvo et al. subsequently confirmed these findings by demonstrating temporolimbic fiber loss, also paradoxically associated with increased regional network efficiency (105). The neurobiological correlates of these findings are yet to be fully defined, but they likely represent regional changes in connectivity strength, with alterations in the natural balance of associations between regions. By changing the strength with which some regions are associated with each other, the conformation of the limbic network can lead to a relative regional strengthening of connections, even though there is overall fiber loss associated with epilepsy.

MRI and postoperative seizure outcome

Refractory TLE is clearly a systems disorder without a circumscribed brain structural alteration. Understanding how focal and networked structural alterations are related to unsuccessful surgery for refractory temporal lobe seizures is an important research objective and is the focus of this section. This paper is concerned with postoperative seizure outcome, and the use of preoperative imaging to predict postoperative cognitive (106,107), psychiatric (108) and visual (109,110) outcome is beyond the scope of this article. Moreover, this paper is concerned with quantitative MRI and DTI, and the reader is referred to other articles for reviews on other imaging modalities (e.g., functional, metabolic) in refractory focal epilepsy (111,112).

Postoperative MRI: extent of resection

It is naturally important to initially consider whether the amount of potentially epileptogenic tissue resected has a significant influence on postoperative seizure control. Although there are some papers that report such a relationship (113-115), there are others that do not (116-118). It may be that inconsistencies may relate to methodological differences between studies. For example, some studies have (I) compared outcome between patients who underwent ATL and those who underwent selective amygdalohippocampectomy (AH) (119); (II) determined resection length on MRI (114); (III) rated whether a structure was completely, partially or not resected on MRI (115,117,118); or (IV) made assessment of resection parameters during surgery (113,116). Differences in outcome classification, including the classification scale and, more likely, time to postoperative assessment may also explain different findings. A review paper on temporal lobe resection and outcome reported that the extent of resection does not necessarily lead to improved postoperative seizure outcome, that patients with significant hippocampal and amygdaloid remnants may experience excellent postoperative seizure outcomes, and that AH and ATL do not differ in rates of seizure freedom (120). On the contrary, a meta-analysis revealed that ATL was significantly more likely to result in seizure freedom relative to AH (119). However, it must be concluded that class I evidence for the extent and type of resection being related to postoperative seizure outcome is rare (120). What is included in the resection may have a significant influence on outcome; two previous studies reported that the extent of entorhinal and parahippocampal resection is significantly related to outcome (121,122). Such findings indicate that preoperative identification of entorhinal and parahippocampal abnormalities may potentially identify a particular subtype of refractory TLE that is less responsive to circumscribed resections (101).

Preoperative volumetric MRI

Given that the hippocampus is considered the primary seizure generator in TLE (123)—or a primary node in an epileptiform network—it is logical to start by rigorously examining the morphology and pathology of the hippocampus using quantitative MRI. Many preoperative evaluation programs include quantitative analysis of hippocampal volume, and there have been some studies that have analyzed such data with respect to postoperative seizure outcome. Using these approaches, Jack et al. (124) reported that a larger hippocampal volume ipsilateral to resection was related to a poorer outcome in a mixed cohort of patients with TLE with and without HS, and Jutila et al. (125) reported that hippocampal volume reduction of at least one standard deviation from the mean of controls was related to improved seizure outcome. Using a classification of HA or no HA based on volumetry, patients achieving an excellent postoperative seizure outcome were significantly more likely to have HA based on a simple classification analysis (126-128). Patients with bilateral or no HA based on volumetry are significantly less likely to attain seizure freedom after surgery relative to patients with clear unilateral HA (10), although satisfactory postoperative outcomes in patients with bilaterally symmetric hippocampal volumes are achievable (129). Conversely, other studies have found no relationship with global hippocampal volume and postoperative seizure control when outcome groups were retrospectively compared (130-132). Although it is generally accepted that the presence of HS leads to improved postoperative seizure control relative to patients with unremarkable MRI, up to 40% of TLE patients with neuroradiological diagnosed unilateral HS—manifested as HA for the majority of patients—will continue to experience postoperative seizures (2). Therefore hippocampal volumetry is unlikely to reliably prospectively stratify individual patients with TLE and HS into probable outcome groups.

Soon after the introduction of increasingly sophisticated quantitative MRI techniques there was the application of automated approaches to relate preoperative hippocampal and extrahippocampal alterations to postoperative outcome in TLE. Studies have reported regional surface (133) and density (134) alterations of the hippocampus contralateral to intended resection in patients with persistent postoperative seizures relative to those rendered seizure free. A previous study reported that 25% of patients with TLE and HS who continued to experience seizures had postoperative epileptiform contributions from the contralateral temporal lobe (135). Bitemporal epileptogenic activity previously undetected during presurgical evaluation could therefore contribute to persistent postoperative seizures in some patients.

In a small group of patients with left TLE and pathologically proven HS, Keller et al. (134) reported that patients with persistent postoperative seizures had significantly reduced grey matter density of the posterior hippocampal region relative to those rendered seizure free using VBM. The authors suggested that given that temporal lobe resections typically leave a posterior hippocampal remnant, it may be that the hippocampal remnant is epileptogenic, remains functionally connected after surgery, and contributes to persistent postoperative seizures. This finding has been recently replicated using a hippocampal surface mapping technique in a larger cohort of patients with TLE and HS, and also confirmed that global hippocampal volume on MRI was not a predictor of postoperative seizure outcome (70). Patients with a posterior mesial temporal lobe seizure onset are more likely to experience persistent postoperative seizures relative to patients with an anterior mesial temporal lobe onset (136,137), and patients with a poor outcome are more likely to have a distribution of neuronal cell loss throughout the anterior-posterior extent of the hippocampus relative to patients with an excellent outcome who had cell loss confined to anterior—and thus resected—regions of the hippocampus (138). However, the finding of posterior HA being preferentially observed in patients with persistent seizures has not been replicated in a study that quantified the volume of the hippocampal head, body and tail in patients with TLE (132), and a previous electrophysiological study indicated that the hippocampal remnant after surgery was the cause of postoperative seizures in only 5% of patients with TLE and HS (135).

Within group comparison morphometric studies, there are reports of increasing extrahippocampal structural alterations in patients with persistent postoperative seizures relative to those rendered seizure free (70,71,90,134,139-141). Extrahippocampal involvement in postoperative seizures has been related to increasing atrophy of the contralateral entorhinal cortex (141), atrophy of the temporopolar and insular cortices (71), atrophy of the parahippocampal region bilaterally (134), alterations in thalamotemporal structure (70), increased whole-brain structural network disruption (90), increased number of anatomically non-specific extrahippocampal abnormalities (139), and generalized grey and white matter atrophy (140), on preoperative MRI. Inferences made from most of these studies are that patients who have an excellent seizure outcome have cerebral alterations mostly restricted to the mesial temporal lobe being resected or disconnected. This suggestion, however, would need to be reconciled with reports of bilateral extrahippocampal structural alterations in patients with well-characterised unilateral TLE and ipsilateral HS who are likely to have standard (up to 60-70% seizure freedom) postoperative seizure outcome rates (97). This has led some to suggest that bilateral temporal and extratemporal structural alterations in patients with TLE and neuroradiological evidence of unilateral HS is common and may not be predictive of poor surgical outcome (142).

It is likely that different findings on the relationship between preoperative brain structure on MRI and postoperative seizure outcome are due to a number of methodological factors. Table 2 presents a breakdown of methodological information across selected studies that have addressed the preoperative neuroanatomical correlates of postoperative seizures using quantitative MRI techniques. From this table it emerges that no single methodological parameter is consistent across studies. In particular, studies differ substantially in terms of (I) the characterization of patients; (II) number of participants studied; (III) surgical approach; (IV) classification of seizure freedom; (V) MRI acquisition; (VI) the area of the brain assessed; and (VII) the morphometric technique employed. In samples of patients with TLE who have not been preselected according to hippocampal pathology (i.e., to include those with no MRI changes, or those with lesional pathology other than HS), HA on MRI ipsilateral to the side of seizure onset is consistently associated with an improved postoperative seizure outcome. However, this is not the case when patients are enrolled into studies on the basis of electrophysiological evidence of TLE and HS, which are the strongest indicators for mesial TLE (mTLE). In these cases, it may be that extrahippocampal or bitemporal alterations may be associated with a poor outcome, or perhaps that a so-far unidentified subtype of mTLE that is less amenable to resective surgery exists.

Table 2. Results from selected volumetric and morphometric MRI studies that have addressed the relationship between preoperative brain structure and postoperative seizure outcome in patients with TLE. mTLE is indicated for samples with electrophysiological evidence of TLE and concomitant ipsilateral HS. Engel’s outcome I refers to all sub-classifications (a-d), unless indicated. ATL does not necessarily refer to en bloc procedure and may refer to AH with additional removal of lateral cortex.

Study	Patients	n*	Surgery	Seizure freedom	Controls	MRI system	MRI sequence	Morphometric technique	Main findings
Bernhardt et al. (71)	TLE with and without HS	47 (TLE without HS; 22 underwent surgery); 58 (TLE with HS; 40 underwent surgery)	AH & ATL	Engel’s Ia	48	1.5 T Philips	3D T1-weighted (FFE)	Cortical thickness (CLASP)	No differences between outcome groups in brain regions shown to be different between patients and controls; some regional cortical differences between outcome groups when statistics not corrected for multiple comparisons
Bernhardt et al. (141)	TLE with and without HS	62 (TLE without HS); 72 (TLE with HS); 90 patients in total underwent surgery	AH	Engel’s I	47	1.5 T Philips	3D T1-weighted (FFE)	Surface shape analysis of medial temporal lobe structures (SPHARM-PDM)	Increasing atrophy of contralateral entorhinal cortex related to PS
Cascino et al. (126)	TLE	30 (21 underwent surgery)	AH, ATL	Engel’s I	–	n/p	n/p	Manual volumetric measurements of the hippocampus	Improved outcome associated with HA
Cascino et al. (127)	TLE	165	AH, ATL	Engel’s I	–	1.5 T GE	n/p	Manual volumetric measurements of the hippocampus	Improved outcome associated with HA
Feis et al. (143)	mTLE with HS (left only)	49	AH	ILAE I-II	–	3 T Siemens	3D T1-weighted (MPRAGE)	Support vector classification applied to VBM (SPM 8) processed images; white matter voxels only	Outcome dependent on gender; correct classification 94% males, 96% females; differences between outcome groups corresponding to areas approximately located in cingulum, fronto-occipital fasciculus and caudate (males) and longitudinal fasciculi (females)
Goh et al. (132)	mTLE with HS	86	AH	Engel’s I	–	3 T Siemens	3D T1-weighted (MPRAGE)	Hippocampal volume (stereology); volumes parcellated into hippocampal head, body and tail	No significant relationship between outcome and volumes
Jack et al. (124)	TLE	50	ATL	Engel’s I & IIa-b (equivalent)	50	1.5 T GE	T1-weighted coronal sections	Manual volumetric measurements of the hippocampus	Volume of operated, but not non-operated, hippocampus significantly related to outcome; concomitant reduced volume and EEG lateralisation required for good outcome
Jutila et al. (125)	TLE	140; unilateral HA in 21%; normal MRI in 34%; various combinations of pathology in remainder	AH, ATL	Engel’s I-II	20	Multiple scanners. Most scanned using a 1.5 T Siemens	3D T1-weighted (MPRAGE)	Manual volumetric measurements of the hippocampus and amygdala	Ipsilateral hippocampal volume reduction over 1 SD from controls related to improved outcome
Keller et al. (134)	mTLE with HS (left only)	22	AH, ATL	Engel’s I	77	1.5 T GE	3D T1-weighted (SPGR)	VBM (SPM 99)	Bilateral posterior medial temporal lobe atrophy in patients with PS relative to those rendered SF
Keller et al. (70)	mTLE with HS	115 (87 underwent surgery)	AH	ILAE I	80	3 T Siemens	3D T1-weighted (MPRAGE)	(I) Cortical thickness (Freesurfer)	Bilateral thalamic and contralateral hippocampal region-specific atrophy in patients with PS relative to those rendered SF; Ipsilateral posterior HA at uncorrected statistical levels
								(I) automated subcortical volumetry (FSL)
								(II) subcortical surface shape (FSL)
Knowlton et al. (128)	TLE	25	ATL	Engel I	19	1.5 T GE	3D T1-weighted (SPGR)	Manual volumetric measurements of the hippocampus and hippocampal T2 relaxometry	Improved outcome associated with HA and increased hippocampal T2 data
Lin et al. (133)	mTLE with HS	40	ATL	ILAE I-II / Engel’s Ia-Ib	–	1.5 T GE	3D T1-weighted (SPGR)	Surface shape analysis of the hippocampi (radial atrophy from hippocampal centreline to surface)	More diffuse ipsilateral and region-specific contralateral HA in those with PS relative to those SF
Mueller et al. (131)	TLE	207 (186 with HS,	AH, ATL, MTL	ILAE I	–	n/p: likely to be different across centers	n/p: likely 3D T1-weighted across centers	Manual volumetric measurements of the hippocampus, amygdala, parahippocampal gyrus, fusiform gyrus, combined inferior and middle temporal gyrus, and superior temporal gyrus	No significant relationship between outcome and volumes
		21 without HS)**
Quigg et al. (130)	mTLE with HS	40	ATL	Engel’s I	23	1.5 T Siemens	3D T1-weighted (MPRAGE)	Manual volumetric measurements of the hippocampus and amygdala	Surgical outcome independent of hippocampal volume asymmetry
Sisodiya et al. (139)	mTLE with HS***	27	ATL	Engel’s Ia	33	1.5 T GE	3D T1-weighted (SPGR)	Manual analysis of regional distribution of brain volume (extrahippocampal structural abnormalities)	Extrahippocampal structural abnormalities in 14, 10 of whom experienced PS; 11 of 13 patients without abnormalities became SF
Yasuda et al. (140)	mTLE with HS	67	AH, ATL	Engel’s Ia	69	2 T Elscint	3D T1-weighted	VBM (SPM 2) and manual hippocampal volumetry	Widespread structural alterations in patients with poor outcome (Engel’s IIb-IV) relative to patients rendered SF or those with worthwhile improvement (Engel’s Ib-IIa)

*, unless indicated, all patients underwent surgery; **, from three surgical centres; patients without HS showed evidence of focal cortical dysplasia [6], ganglioglioma [5], no histological alterations [2], gliosis [4], cavernoma [2], hamartoma [1] or microdysgenesis [1]; ***, two patients found to additionally have dual pathology (HS with subependymal heterotopia). mTLE, mesial temporal lobe epilepsy; AH, amygdalohippocampectomy; ATL, anterior temporal lobectomy; CLASP, Constrained Laplacian Anatomic Segmentation using Proximity algorithm; FFE, fast field echo; GE, General Electric; HA, hippocampal atrophy; MPRAGE, magnetization-prepared rapid gradient-echo; MTL, mesial temporal lesionectomy; n/p, not provided; PS, persistent postoperative seizures; SD, standard deviation; SF, seizure free; SPGR, spoiled gradient echo; SPHARM-PDM, spherical harmonics with a point distribution model; VBM, voxel-based morphometry.

DTI: connectivity and networks

There is abundant evidence to suggest that voxel-based scalars of water diffusion, as well as DTI fiber tractography, are abnormal in patients with TLE, notably within areas that are functionally or anatomically connected with the medial temporal regions. Nonetheless, the relevance of these findings towards clinical outcomes is not yet fully defined. A growing body of evidence suggests that extrahippocampal abnormalities are indeed a crucial determinant of neuropsychological performance. Winston et al. demonstrated that white matter integrity in a frontoparietal network and in the contralateral temporal lobe is associated with working memory performance (144). McDonald et al. reported that structural compromise to multiple DTI streamlines tracts is associated with memory and language impairments in patients with TLE (145) and memory performance is better predicted when DTI data regarding white matter integrity is combined with gray matter information from conventional morphometric measures (146). Interestingly, the one to one relationship between memory type deficits and anatomical pathways has not been fully determined. Certainly, medial temporal outflow pathways such as the fornix may play an important role (147), but the overall distribution across extratemporal regions is likely more widespread, or perhaps variable across individuals.

The relationship between seizure control and white matter integrity demonstrated by DTI is starting to be explored in a systematic way. Gonçalves Pereira demonstrated that hippocampal MD asymmetry indices are higher in patients with optimal surgical outcome (148). The relationship between extrahippocampal abnormalities and outcome is also likely important, as structural connectivity demonstrates a reorganization of strong temporal-extratemporal connections in patients with suboptimal outcome (149). While these are preliminary findings, there appears to be a growing body of evidence to suggest that white matter organization supports the clinical profile of epilepsy and seizure control.

Future work

The development of imaging prognostic classifiers that can stratify individual patients according to likely outcome is an important future direction. It may be that quantitative MRI alone cannot achieve this, particularly using the current resolution of clinical MRI systems, and that multimodal imaging (combining structure, function and metabolism) will be more informative. There is evidence indicating that temporal lobe (150,151) and extratemporal lobe (152,153) fluorine-18 fluorodeoxyglucose positron emission tomography (FDG-PET) hypometabolism is an indicator for postoperative seizure outcome. Magnetic resonance spectroscopy (MRS) has been shown to have excellent lateralising value in patients with TLE, particularly in conjunction with hippocampal volumetry (154), shows promise in predicting pharmacoresistance (155), but has mixed evidence for the ability to predict postoperative outcome in unilateral TLE (128,156-158). Resting-state functional MRI may also offer promise for in the prediction of postoperative outcome (159). Whether these multimodal quantitative imaging techniques can be used alongside quantitative MRI and DTI methods for individual prognostic classification remains to be seen.

Classical HS is characterized by preferential neuron loss and gliosis in CA1, and also to a lesser extent of CA4 and CA3, with relative resistance of CA2 neurons (160,161). However, over half of patients with well-characterized refractory mesial TLE have neuronal loss throughout all CA sectors (162). It has been demonstrated that patients with classical and total HS, who make up the vast majority of patients with TLE and HS, have a greater chance of seizure freedom after surgery relative to patients with no histopathological evidence of HS, or atypical patterns of HS manifested as circumscribed loss of neurons in CA1 or CA4 (162,163). Patients with atypical patterns of HS may be considered to have a particular subtype of TLE that is resistant to conventional resective surgery. An important future quantitative MRI research endeavor would therefore be an attempt to identify atypical patterns of HS on preoperative MRI as this may suggest a potential poor postoperative prognosis in those patients. To our knowledge, there have been no quantitative MRI studies attempting to correlate atypical patterns of HS prior to surgery with postoperative outcome, despite some progress in the identification of hippocampal subfields using high-field MRI (164) and probabilistic mapping (165).

Regarding white matter networks and TLE, there are undoubtedly several questions that remain to be answered, including whether (I) abnormal networks are the same for all patients with TLE; (II) there are crucial areas within the network where abnormalities lead to surgical refractoriness; (III) pharmacological intractability is also a function of abnormal networks and can be tracked over the time to provide a measure of long-term trajectory and guide management decisions; (IV) abnormalities are reversible after successful treatment; and (V) white matter topography can be used as an individualized measurement of clinical course. It is conceivable that several of these questions may be adequately answered by DTI analyses. An example of possible anatomical pathways of seizure spread is demonstrated in Figure 3. By examining the structural properties of these pathways, alone or in combination, structural connectivity may provide information regarding the architectural variability across individuals with epilepsy. Importantly, this information may be used to summarize the degree and extent of network abnormalities and possibly be used to provide individualized measures of operative outcome.

Figure 3.

This diagram illustrates anatomical pathways of possible seizure spread in patients with TLE. The lines represent principal anatomical connectivity routes connecting the hippocampus and amygdala to frontal, orbitofrontal and cingulate regions (green), medial temporal and occipital regions (cyan), insula and subcortical grey matter structures (blue), lateral and polar temporal regions (yellow) and contralateral medial temporal structure (black). These pathways may be preferentially affected in each individual with TLE. By examining architectural changes in these specific routes, alone or in combination, future studies employing the structural connectome may disclose abnormalities associated with the extent of epileptogenic networks and their relationship with surgical outcome. TLE, temporal lobe epilepsy.

Conclusions

Refractory TLE is characterized by networked brain structural alterations that exist primarily in limbic and paralimbic regions, which have been reproducibly demonstrated across studies using quantitative MRI and DTI techniques. Some of these structural and connectivity alterations may have significance for why some patients respond well to resective surgery and others do not. An important future goal is to use quantitative imaging techniques to identify a particular subtype of TLE that is particularly resistant to conventional temporal lobe surgery. Whether this information can be applied to stratify individual patients according to predicted outcome remains to be determined.